Traditional pond crab culture mainly relies on fishermen to estimate the total bait based on experience, and feed bait by manual punting, which has low bait utilization rate and high labor intensity. Because river crabs have territorial awareness and small moving range, the distribution of river crabs in the pond is uneven, thus the scientific and accurate feeding is required for the crab culture. The existing feeding operation mode of river crab culture is extensive, which can not meet the needs of efficient ecological culture of river crab. In order to grasp the growth law of river crabs and feed more scientifically and effectively, a precise feeding system based on river crab growth model was designed. The grey correlation analysis method was adopted in the growth model of the river crab to determine the environmental factors that have the greatest impact on the growth and development of the river crab. Based on the traditional aquatic biological growth model, environmental factors were added to improve the river crab growth model, which was optimized and fitted from the linear and exponential perspectives. The GA-BP neural network was used to train the accurate feeding prediction model, and the optimal environmental impact factor value was calculated by inputting environmental parameters such as water temperature, dissolved oxygen content, and pH value. Then the total weight of the crabs was obtained according to the growth model, breeding density and breeding area of the river crab. Combined with the survival rate and feeding rate of river crab during the growth period, the total bait weight can be determined. Finally, according to the actual distribution density of crabs and water quality parameters, the bait distribution coefficient of each area in the pond was determined, and the total bait was allocated to each area of the pond scientifically. The simulation results showed that the determination coefficient R2 of predicted total bait weight was 0.990, and the fitting effect of the prediction model was good. Through pond feeding experiments, the results showed that based on the total bait determination by using the growth model of river crab, the pond area that could be accurately fed by the automatic feeding boat was 5.33hm2, saving the labor cost of three farmers. Compared with the preset feeding density for each area of the pond, the average absolute error of the actual feeding density performed by the feeding boat was 0.32g/m2 and the average relative error was 3.90%. In addition, the feeding weight can be adjusted timely by the system according to the changes of the environmental parameters and the feedback from feeding table, which was conducive to saving bait, cultivating large crabs, increasing crab production, improving breeding efficiency, and promoting cost-effective development of crab culture.